Good Papers: read the good ones

Quotienting Markov chains by their peripheral invariant subspace separates persistent regime profiles from transient dynamics for stable policy evaluation.

Yang Xu, Vaneet Aggarwal

Atlanta Poster Session 6, Fri, Dec 11, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

78%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Columbia Learning theory

A mathematical theory of balancing relational generalization and memorization

A theory of transitive inference with exceptions shows relational generalization and memorization depend on representational geometry, with pretrained language models exhibiting predicted systematic errors.

Luke Cheng, Samuel Lippl

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

11/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 11 of 20 reviewers recommend it

lenient 4/5

medium 5/10

strict 2/5

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Cambridge ETH Zurich ETH Zürich Learning theory

Generating in the Limit with Infinitely Many Hallucinations

Language generation in the limit is recast as recall-precision trade-offs, showing that allowing infinitely many vanishing-frequency hallucinations can strictly increase recall when adversaries withhold target portions.

Irene Strauss, Alexandra Butoi, Ryan Cotterell

Sydney Poster Session 6, Thu, Dec 10, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026SpotlightHSLU Learning theory

Nearly Optimal Robust Covariance and Scatter Matrix Estimation Beyond Gaussians

A polynomial-time algorithm robustly estimates elliptical scatter matrices with nearly optimal samples and error beyond Gaussians.

Gleb Novikov

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026RIKEN AIP The Univ The University of Osaka / RIKEN Learning theory

Information-Theoretic Generalization Bounds for Sequential Decision Making

A sequential supersample framework bounds sequential decision-making generalization via roundwise mutual information and faster Bernstein rates, applying to online learning and bandits.

Futoshi Futami, Masahiro Fujisawa

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 8 of 20 reviewers recommend it

lenient 2/5

medium 4/10

strict 2/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Washington U Southern California Learning theory

Adaptive Calibration in Non-Stationary Environments

Online prediction algorithms achieve calibration error adapting to non-stationarity via $\tilde O(\min\{\sqrt{T}+(TC)^{1/3},\sqrt{KT}\})$ bounds, smoothly interpolating between i.i.d. and adversarial settings.

Junyan Liu, Haipeng Luo, Lillian Ratliff

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

9/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

88%Must read

?Must readVote to see the score

NeurIPS 2026National University of Singapore U Cambridge Learning theory

Beyond Worst-Case Coreset Bounds for $k$-Clustering via Determinantal Sampling

Determinantal sampling builds smaller k-clustering coresets with sub-quadratic ε dependence under mild data assumptions, breaking worst-case bounds.

Diptarka Chakraborty, Satyaki Mukherjee, Gaurav Vallabhdas Revankar, Hoang Son Tran

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

– ReadersNo votes yet

15/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026KAIST Seoul National Learning theory

Logistic Bandits with $\tilde{O}(\sqrt{dT})$ Regret without Context Diversity Assumptions

SupSplitLog achieves near-optimal logistic bandit regret without context diversity assumptions via sample splitting and Newton-type corrections.

Seoungbin Bae, Dabeen Lee

Sydney Poster Session 5, Thu, Dec 10, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Sapienza University of Rome London School of Economics Aarhus Learning theory

Profit Maximization in Bilateral Trade against a Smooth Adversary

A profit-maximizing broker achieves tight O√T regret against smooth adversaries in bilateral trade via continuity and hierarchical nets.

Simone Di Gregorio, Paul Duetting, Federico Fusco, Chris Schwiegelshohn

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 8 of 20 reviewers recommend it

lenient 2/5

medium 4/10

strict 2/5

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026BCAM Basque Center for Applied Mathem U Michigan Learning theory

Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification

A selective-classification method estimates label-noise transition matrices with finite-sample guarantees while bypassing fragile class-posterior estimation.

Xabier de Juan, Santiago Mazuelas, Yilun Zhu, Clay Scott

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 10 of 20 reviewers recommend it

lenient 5/5

medium 5/10

strict 0/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Johns Hopkins Learning theory

Tree Search With Predictions

No algorithm achieves O(log η) search on general trees via distance predictions, but O(k log η) queries work for trees of pathwidth k with optimal complexity.

Michael Dinitz, Bob Dong

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

9/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

76%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Tel Aviv U Iowa Tel Aviv University / Google Learning theory

Near-Optimal Stochastic Linear Bandits with Delay

Stochastic linear bandits with delayed feedback yield near-optimal, dimension-free additive penalties for loss-independent delays but dimension-dependent penalties for loss-dependent delays, unlike multi-armed bandits.

Ofir Schlisselberg, Mengxiao Zhang, Yishay Mansour

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

– ReadersNo votes yet

10/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 10 of 20 reviewers recommend it

lenient 2/5

medium 4/10

strict 4/5

74%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Amazon IMPA Nanyang Technological Learning theory

Mean Testing under Truncation beyond Gaussian

Under truncation hiding an ε-fraction of mass, mean testing faces a bias floor of order ν ε^{1−1/p}; above it a second-order test achieves near-optimal sample complexity, while median regularity restores classical √d testing rates.

Yuhao Wang, Roberto I Oliveira, Themis Gouleakis

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

9/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026U Massachusetts Amherst U Massachusetts at Amherst State University of New York at University at Buffalo Learning theory

Intrinsic Riemannian Cross-covariance for Manifold-valued Random Objects

Intrinsic Riemannian cross-covariance defines manifold-valued covariance via parallel transport to a common tangent space, yielding coordinate-independent second-order descriptors with Euclidean-like properties and verified asymptotic behavior.

Carlos Soto, Cheng Wang, Yujing Huang, Xiaoyu Chen

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Georgia Institute of Technology Georgia Tech Learning theory

Learning Theory of Transformers: Local-to-Global Approximation via Softmax Partition of Unity

Transformers approximate α-Hölder functions via softmax partition of unity with two encoder blocks, achieving near minimax-optimal generalization rates.

Zhongjie Shi, Wenjing Liao

Atlanta Poster Session 4, Thu, Dec 10, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 7 of 20 reviewers recommend it

lenient 2/5

medium 4/10

strict 1/5

71%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Harvard Harvard University/Google Resear Learning theory

Learning to Persuade a Biased Receiver

Proposes safe exploration to learn a receiver's unknown belief bias via signaling, achieving optimal O(log log T) regret by exploiting asymmetric probing costs.

Yuqi Pan, Sadie Zhao, Milind Tambe, Yiling Chen

Atlanta Poster Session 5, Fri, Dec 11, 10:00 AM–1:00 PM, Hall C1 · Published 2026

– ReadersNo votes yet

7/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

AI panel: 7 of 20 reviewers recommend it

lenient 2/5

medium 3/10

strict 2/5

80%Must read

?Must readVote to see the score

NeurIPS 2026SpotlightKAIST AI UNIST Ewha Womans Learning theory

Deep Barycentric Regression for Optimal Transport Map Estimation and its Statistical Optimality

BROT estimates optimal transport maps via barycentric regression with deep networks, achieving minimax optimal convergence rates under Lipschitz conditions with stable training.

Kunwoong Kim, Insung Kong, Yongdai Kim

Sydney Poster Session 3, Wed, Dec 9, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

– ReadersNo votes yet

12/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

72%Highly rated

?Highly ratedVote to see the score

NeurIPS 2026Carnegie Mellon Learning theory

Persuasive Prediction via Decision Calibration

Persuasive prediction learns decision-calibrated predictors from data without common priors, matching Bayesian persuasion utility with efficient algorithms.

Jingwu Tang, Jiahao Zhang, Fei Fang, Steven Wu

Atlanta Poster Session 2, Wed, Dec 9, 4:30 PM–7:30 PM, Hall C1 · Published 2026

– ReadersNo votes yet

8/20 AI panelreviewers recommend it

Readers and the AI panel: vote on this paper to see what they said.

Only vote on papers you've read. Sign in with GitHub to vote.

Show 20 more papers